x
Eliciting base models with simple unsupervised techniques — LessWrong